Small Size Team
Michael Sokolsky, Dinesh Govindaraju, Jennifer Lin, Betsy Ricker,
Small-size robot soccer is a game between two autonomous teams of 5 robots
with an orange golf ball. The robots compete in a game of soccer with
rules dictated by a human referee. Referee signals are transmitted
to each team computer via the referee box. Small-size differs from the
other leagues in a number of ways. The robots, which are built by the teams, must
fit within a 180cm cylinder that is 15cm tall. The robots must also be
relatively concave and have no holes that swallow the ball by more
than 20% of its surface area when projected onto the ground (ie. about
1cm of depth). Overhead cameras are allowed in addition to on-robot
sensing. Additionally, off-field computers are allowed to 'simplify'
the distributed computation challenges. These rules allow teams to
focus more on the behavior and teamwork part of the robot soccer
problem rather than on single robot issues such as localization, which
are an ever present problem in leagues such as the Sony AIBO
league. As such, small-size league games are amongst the fastest and most
dynamic within the leagues that make up RoboCup. Watch some of our
team videos (see below) and see for yourself.
Carnegie Mellon has participated in the small-size league since its
inception in 1997. We have won the competition twice, have been
semi-finalists, and quarter finalists twice. Our team has a number of
unique features, including:
- A unique hierarchical control system designed for its control
response properties rather than adherence to a particular deliberation
vs. reactive philosophy
- An adaptive strategy engine that adapts team behavior based on its performance
as the game progresses.
- A play based representation for human readable representation
of team behavior under different situations.
- A tactics layer for encoding of individual robot skills. This
includes a two layered finite-state-machine for controlling actions,
facilities for rapidly evaluating different objective
functions, obstacle free near-optimal navigation, and near-optimal
trapezoidal motion control.
- Very fast randomized path planning (~2ms per plan) for
efficient robot navigation
- A very fast color vision library CMVision, which is also used
on our CMPack, Sony AIBO team
- A robust high-level vision system that includes fast
calibration programs for camera geometry calibration, and color
- A realistic 3D simulator that
incorporates a realistic physics engine with reconfigurable robot parameters.
We have a number of Open Source releases, of both software and
hardware, that are available to everyone. These items are available
under the GNU LGPL license. We just ask that if you use our releases that
you cite our relevant papers appropriately (see our publication
Go to the CORAL
Download page for our full list of downloads.
- SSL-Vision the shared Small Size League Vision system, developed in part by CMU (also see the paper here).
- CMVision the
fast color vision library used by CMDragons, now part of SSL-Vision.
2002 RoboCup code base can be found here.
We have a collection of videos and pictures in the following
World Champions of the RoboCup'98 Small-Size Robot League
World Champions of the RoboCup'97 Small-Size Robot League
Our small-size publications can be found here: